Items, such as products available via an electronic commerce system, can become popular with little warning. It can be difficult to predict which items achieve popularity or success before the items actually achieve popularity or success. Prediction of item trends that consider backward looking sales metrics can result in detection of trends associated with an item later than is desired.